2 * Copyright (c) 2007 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.30 2008/02/10 09:51:01 dillon Exp $
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/fcntl.h>
41 #include <sys/namecache.h>
42 #include <sys/vnode.h>
43 #include <sys/lockf.h>
44 #include <sys/event.h>
46 #include <sys/dirent.h>
47 #include <vm/vm_extern.h>
48 #include <vfs/fifofs/fifo.h>
54 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
55 static int hammer_vop_fsync(struct vop_fsync_args *);
56 static int hammer_vop_read(struct vop_read_args *);
57 static int hammer_vop_write(struct vop_write_args *);
58 static int hammer_vop_access(struct vop_access_args *);
59 static int hammer_vop_advlock(struct vop_advlock_args *);
60 static int hammer_vop_close(struct vop_close_args *);
61 static int hammer_vop_ncreate(struct vop_ncreate_args *);
62 static int hammer_vop_getattr(struct vop_getattr_args *);
63 static int hammer_vop_nresolve(struct vop_nresolve_args *);
64 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
65 static int hammer_vop_nlink(struct vop_nlink_args *);
66 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
67 static int hammer_vop_nmknod(struct vop_nmknod_args *);
68 static int hammer_vop_open(struct vop_open_args *);
69 static int hammer_vop_pathconf(struct vop_pathconf_args *);
70 static int hammer_vop_print(struct vop_print_args *);
71 static int hammer_vop_readdir(struct vop_readdir_args *);
72 static int hammer_vop_readlink(struct vop_readlink_args *);
73 static int hammer_vop_nremove(struct vop_nremove_args *);
74 static int hammer_vop_nrename(struct vop_nrename_args *);
75 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
76 static int hammer_vop_setattr(struct vop_setattr_args *);
77 static int hammer_vop_strategy(struct vop_strategy_args *);
78 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
79 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
80 static int hammer_vop_ioctl(struct vop_ioctl_args *);
81 static int hammer_vop_mountctl(struct vop_mountctl_args *);
83 static int hammer_vop_fifoclose (struct vop_close_args *);
84 static int hammer_vop_fiforead (struct vop_read_args *);
85 static int hammer_vop_fifowrite (struct vop_write_args *);
87 static int hammer_vop_specclose (struct vop_close_args *);
88 static int hammer_vop_specread (struct vop_read_args *);
89 static int hammer_vop_specwrite (struct vop_write_args *);
91 struct vop_ops hammer_vnode_vops = {
92 .vop_default = vop_defaultop,
93 .vop_fsync = hammer_vop_fsync,
94 .vop_getpages = vop_stdgetpages,
95 .vop_putpages = vop_stdputpages,
96 .vop_read = hammer_vop_read,
97 .vop_write = hammer_vop_write,
98 .vop_access = hammer_vop_access,
99 .vop_advlock = hammer_vop_advlock,
100 .vop_close = hammer_vop_close,
101 .vop_ncreate = hammer_vop_ncreate,
102 .vop_getattr = hammer_vop_getattr,
103 .vop_inactive = hammer_vop_inactive,
104 .vop_reclaim = hammer_vop_reclaim,
105 .vop_nresolve = hammer_vop_nresolve,
106 .vop_nlookupdotdot = hammer_vop_nlookupdotdot,
107 .vop_nlink = hammer_vop_nlink,
108 .vop_nmkdir = hammer_vop_nmkdir,
109 .vop_nmknod = hammer_vop_nmknod,
110 .vop_open = hammer_vop_open,
111 .vop_pathconf = hammer_vop_pathconf,
112 .vop_print = hammer_vop_print,
113 .vop_readdir = hammer_vop_readdir,
114 .vop_readlink = hammer_vop_readlink,
115 .vop_nremove = hammer_vop_nremove,
116 .vop_nrename = hammer_vop_nrename,
117 .vop_nrmdir = hammer_vop_nrmdir,
118 .vop_setattr = hammer_vop_setattr,
119 .vop_strategy = hammer_vop_strategy,
120 .vop_nsymlink = hammer_vop_nsymlink,
121 .vop_nwhiteout = hammer_vop_nwhiteout,
122 .vop_ioctl = hammer_vop_ioctl,
123 .vop_mountctl = hammer_vop_mountctl
126 struct vop_ops hammer_spec_vops = {
127 .vop_default = spec_vnoperate,
128 .vop_fsync = hammer_vop_fsync,
129 .vop_read = hammer_vop_specread,
130 .vop_write = hammer_vop_specwrite,
131 .vop_access = hammer_vop_access,
132 .vop_close = hammer_vop_specclose,
133 .vop_getattr = hammer_vop_getattr,
134 .vop_inactive = hammer_vop_inactive,
135 .vop_reclaim = hammer_vop_reclaim,
136 .vop_setattr = hammer_vop_setattr
139 struct vop_ops hammer_fifo_vops = {
140 .vop_default = fifo_vnoperate,
141 .vop_fsync = hammer_vop_fsync,
142 .vop_read = hammer_vop_fiforead,
143 .vop_write = hammer_vop_fifowrite,
144 .vop_access = hammer_vop_access,
145 .vop_close = hammer_vop_fifoclose,
146 .vop_getattr = hammer_vop_getattr,
147 .vop_inactive = hammer_vop_inactive,
148 .vop_reclaim = hammer_vop_reclaim,
149 .vop_setattr = hammer_vop_setattr
152 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
153 struct ucred *cred, int flags);
154 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
155 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
160 hammer_vop_vnoperate(struct vop_generic_args *)
162 return (VOCALL(&hammer_vnode_vops, ap));
167 * hammer_vop_fsync { vp, waitfor }
171 hammer_vop_fsync(struct vop_fsync_args *ap)
177 error = hammer_sync_inode(ip, ap->a_waitfor, 0);
182 * hammer_vop_read { vp, uio, ioflag, cred }
186 hammer_vop_read(struct vop_read_args *ap)
188 struct hammer_transaction trans;
197 if (ap->a_vp->v_type != VREG)
201 seqcount = ap->a_ioflag >> 16;
203 hammer_start_transaction(&trans, ip->hmp);
206 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
209 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
210 offset = uio->uio_offset & HAMMER_BUFMASK;
212 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
213 uio->uio_offset - offset, HAMMER_BUFSIZE,
214 MAXBSIZE, seqcount, &bp);
216 error = bread(ap->a_vp, uio->uio_offset - offset,
217 HAMMER_BUFSIZE, &bp);
222 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
223 n = HAMMER_BUFSIZE - offset;
224 if (n > uio->uio_resid)
226 if (n > ip->ino_rec.ino_size - uio->uio_offset)
227 n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
228 error = uiomove((char *)bp->b_data + offset, n, uio);
233 if ((ip->flags & HAMMER_INODE_RO) == 0 &&
234 (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
235 ip->ino_rec.ino_atime = trans.tid;
236 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
240 hammer_commit_transaction(&trans);
245 * hammer_vop_write { vp, uio, ioflag, cred }
249 hammer_vop_write(struct vop_write_args *ap)
251 struct hammer_transaction trans;
252 struct hammer_inode *ip;
260 if (ap->a_vp->v_type != VREG)
265 if (ip->flags & HAMMER_INODE_RO)
269 * Create a transaction to cover the operations we perform.
271 hammer_start_transaction(&trans, ip->hmp);
277 if (ap->a_ioflag & IO_APPEND)
278 uio->uio_offset = ip->ino_rec.ino_size;
281 * Check for illegal write offsets. Valid range is 0...2^63-1
283 if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
284 hammer_commit_transaction(&trans);
289 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
291 while (uio->uio_resid > 0) {
294 offset = uio->uio_offset & HAMMER_BUFMASK;
295 n = HAMMER_BUFSIZE - offset;
296 if (n > uio->uio_resid)
298 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
299 vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
303 if (uio->uio_segflg == UIO_NOCOPY) {
305 * Issuing a write with the same data backing the
306 * buffer. Instantiate the buffer to collect the
307 * backing vm pages, then read-in any missing bits.
309 * This case is used by vop_stdputpages().
311 bp = getblk(ap->a_vp, uio->uio_offset - offset,
312 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
313 if ((bp->b_flags & B_CACHE) == 0) {
315 error = bread(ap->a_vp,
316 uio->uio_offset - offset,
317 HAMMER_BUFSIZE, &bp);
319 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
321 * entirely overwrite the buffer
323 bp = getblk(ap->a_vp, uio->uio_offset - offset,
324 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
325 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
329 bp = getblk(ap->a_vp, uio->uio_offset - offset,
330 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
334 * Partial overwrite, read in any missing bits then
335 * replace the portion being written.
337 error = bread(ap->a_vp, uio->uio_offset - offset,
338 HAMMER_BUFSIZE, &bp);
343 error = uiomove((char *)bp->b_data + offset, n, uio);
346 * If we screwed up we have to undo any VM size changes we
352 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
357 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
358 if (ip->ino_rec.ino_size < uio->uio_offset) {
359 ip->ino_rec.ino_size = uio->uio_offset;
360 flags = HAMMER_INODE_RDIRTY;
361 vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
365 ip->ino_rec.ino_mtime = trans.tid;
366 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
367 hammer_modify_inode(&trans, ip, flags);
370 * The file write must be tagged with the same TID as the
371 * inode, for consistency in case the inode changed size.
372 * This guarantees the on-disk data records will have a
373 * TID <= the inode TID representing the size change.
375 * If a prior write has not yet flushed, retain its TID.
378 bp->b_tid = ip->last_tid;
380 if (ap->a_ioflag & IO_SYNC) {
382 } else if (ap->a_ioflag & IO_DIRECT) {
389 hammer_abort_transaction(&trans);
391 hammer_commit_transaction(&trans);
396 * hammer_vop_access { vp, mode, cred }
400 hammer_vop_access(struct vop_access_args *ap)
402 struct hammer_inode *ip = VTOI(ap->a_vp);
407 uid = hammer_to_unix_xid(&ip->ino_data.uid);
408 gid = hammer_to_unix_xid(&ip->ino_data.gid);
410 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
411 ip->ino_data.uflags);
416 * hammer_vop_advlock { vp, id, op, fl, flags }
420 hammer_vop_advlock(struct vop_advlock_args *ap)
422 struct hammer_inode *ip = VTOI(ap->a_vp);
424 return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
428 * hammer_vop_close { vp, fflag }
432 hammer_vop_close(struct vop_close_args *ap)
434 return (vop_stdclose(ap));
438 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
440 * The operating system has already ensured that the directory entry
441 * does not exist and done all appropriate namespace locking.
445 hammer_vop_ncreate(struct vop_ncreate_args *ap)
447 struct hammer_transaction trans;
448 struct hammer_inode *dip;
449 struct hammer_inode *nip;
450 struct nchandle *nch;
454 dip = VTOI(ap->a_dvp);
456 if (dip->flags & HAMMER_INODE_RO)
460 * Create a transaction to cover the operations we perform.
462 hammer_start_transaction(&trans, dip->hmp);
465 * Create a new filesystem object of the requested type. The
466 * returned inode will be referenced but not locked.
469 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
471 kprintf("hammer_create_inode error %d\n", error);
473 hammer_abort_transaction(&trans);
479 * Add the new filesystem object to the directory. This will also
480 * bump the inode's link count.
482 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
484 kprintf("hammer_ip_add_directory error %d\n", error);
490 hammer_rel_inode(nip, 0);
491 hammer_abort_transaction(&trans);
494 hammer_commit_transaction(&trans);
495 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
496 hammer_rel_inode(nip, 0);
498 cache_setunresolved(ap->a_nch);
499 cache_setvp(ap->a_nch, *ap->a_vpp);
506 * hammer_vop_getattr { vp, vap }
510 hammer_vop_getattr(struct vop_getattr_args *ap)
512 struct hammer_inode *ip = VTOI(ap->a_vp);
513 struct vattr *vap = ap->a_vap;
516 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
517 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
522 hammer_itimes(ap->a_vp);
525 vap->va_fsid = ip->hmp->fsid_udev;
526 vap->va_fileid = ip->ino_rec.base.base.obj_id;
527 vap->va_mode = ip->ino_data.mode;
528 vap->va_nlink = ip->ino_rec.ino_nlinks;
529 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
530 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
533 vap->va_size = ip->ino_rec.ino_size;
534 hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
535 hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
536 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
537 vap->va_flags = ip->ino_data.uflags;
538 vap->va_gen = 1; /* hammer inums are unique for all time */
539 vap->va_blocksize = 32768; /* XXX - extract from root volume */
540 vap->va_bytes = ip->ino_rec.ino_size;
541 vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
542 vap->va_filerev = 0; /* XXX */
543 /* mtime uniquely identifies any adjustments made to the file */
544 vap->va_fsmid = ip->ino_rec.ino_mtime;
545 vap->va_uid_uuid = ip->ino_data.uid;
546 vap->va_gid_uuid = ip->ino_data.gid;
547 vap->va_fsid_uuid = ip->hmp->fsid;
548 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
551 switch (ip->ino_rec.base.base.obj_type) {
552 case HAMMER_OBJTYPE_CDEV:
553 case HAMMER_OBJTYPE_BDEV:
554 vap->va_rmajor = ip->ino_data.rmajor;
555 vap->va_rminor = ip->ino_data.rminor;
565 * hammer_vop_nresolve { nch, dvp, cred }
567 * Locate the requested directory entry.
571 hammer_vop_nresolve(struct vop_nresolve_args *ap)
573 struct namecache *ncp;
577 struct hammer_cursor cursor;
578 union hammer_record_ondisk *rec;
588 * Misc initialization, plus handle as-of name extensions. Look for
589 * the '@@' extension. Note that as-of files and directories cannot
592 dip = VTOI(ap->a_dvp);
593 ncp = ap->a_nch->ncp;
594 asof = dip->obj_asof;
598 for (i = 0; i < nlen; ++i) {
599 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
600 asof = hammer_str_to_tid(ncp->nc_name + i + 2);
601 flags |= HAMMER_INODE_RO;
608 * If there is no path component the time extension is relative to
612 ip = hammer_get_inode(dip->hmp, &dip->cache[1], dip->obj_id,
613 asof, flags, &error);
615 error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
616 hammer_rel_inode(ip, 0);
622 cache_setvp(ap->a_nch, vp);
629 * Calculate the namekey and setup the key range for the scan. This
630 * works kinda like a chained hash table where the lower 32 bits
631 * of the namekey synthesize the chain.
633 * The key range is inclusive of both key_beg and key_end.
635 namekey = hammer_directory_namekey(ncp->nc_name, nlen);
637 error = hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
638 cursor.key_beg.obj_id = dip->obj_id;
639 cursor.key_beg.key = namekey;
640 cursor.key_beg.create_tid = 0;
641 cursor.key_beg.delete_tid = 0;
642 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
643 cursor.key_beg.obj_type = 0;
645 cursor.key_end = cursor.key_beg;
646 cursor.key_end.key |= 0xFFFFFFFFULL;
648 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
651 * Scan all matching records (the chain), locate the one matching
652 * the requested path component.
654 * The hammer_ip_*() functions merge in-memory records with on-disk
655 * records for the purposes of the search.
658 error = hammer_ip_first(&cursor, dip);
664 error = hammer_ip_resolve_data(&cursor);
668 if (nlen == rec->entry.base.data_len &&
669 bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
670 obj_id = rec->entry.obj_id;
673 error = hammer_ip_next(&cursor);
675 hammer_done_cursor(&cursor);
677 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
678 obj_id, asof, flags, &error);
680 error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
681 hammer_rel_inode(ip, 0);
687 cache_setvp(ap->a_nch, vp);
690 } else if (error == ENOENT) {
691 cache_setvp(ap->a_nch, NULL);
697 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
699 * Locate the parent directory of a directory vnode.
701 * dvp is referenced but not locked. *vpp must be returned referenced and
702 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
703 * at the root, instead it could indicate that the directory we were in was
706 * NOTE: as-of sequences are not linked into the directory structure. If
707 * we are at the root with a different asof then the mount point, reload
708 * the same directory with the mount point's asof. I'm not sure what this
709 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
710 * get confused, but it hasn't been tested.
714 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
716 struct hammer_inode *dip;
717 struct hammer_inode *ip;
718 int64_t parent_obj_id;
722 dip = VTOI(ap->a_dvp);
723 asof = dip->obj_asof;
724 parent_obj_id = dip->ino_data.parent_obj_id;
726 if (parent_obj_id == 0) {
727 if (dip->obj_id == HAMMER_OBJID_ROOT &&
728 asof != dip->hmp->asof) {
729 parent_obj_id = dip->obj_id;
730 asof = dip->hmp->asof;
731 *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
732 ksnprintf(*ap->a_fakename, 19, "0x%016llx",
740 ip = hammer_get_inode(dip->hmp, &dip->cache[1], parent_obj_id,
741 asof, dip->flags, &error);
746 error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
747 hammer_rel_inode(ip, 0);
752 * hammer_vop_nlink { nch, dvp, vp, cred }
756 hammer_vop_nlink(struct vop_nlink_args *ap)
758 struct hammer_transaction trans;
759 struct hammer_inode *dip;
760 struct hammer_inode *ip;
761 struct nchandle *nch;
765 dip = VTOI(ap->a_dvp);
768 if (dip->flags & HAMMER_INODE_RO)
770 if (ip->flags & HAMMER_INODE_RO)
774 * Create a transaction to cover the operations we perform.
776 hammer_start_transaction(&trans, dip->hmp);
779 * Add the filesystem object to the directory. Note that neither
780 * dip nor ip are referenced or locked, but their vnodes are
781 * referenced. This function will bump the inode's link count.
783 error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
789 hammer_abort_transaction(&trans);
791 cache_setunresolved(nch);
792 cache_setvp(nch, ap->a_vp);
793 hammer_commit_transaction(&trans);
799 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
801 * The operating system has already ensured that the directory entry
802 * does not exist and done all appropriate namespace locking.
806 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
808 struct hammer_transaction trans;
809 struct hammer_inode *dip;
810 struct hammer_inode *nip;
811 struct nchandle *nch;
815 dip = VTOI(ap->a_dvp);
817 if (dip->flags & HAMMER_INODE_RO)
821 * Create a transaction to cover the operations we perform.
823 hammer_start_transaction(&trans, dip->hmp);
826 * Create a new filesystem object of the requested type. The
827 * returned inode will be referenced but not locked.
829 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
831 kprintf("hammer_mkdir error %d\n", error);
833 hammer_abort_transaction(&trans);
839 * Add the new filesystem object to the directory. This will also
840 * bump the inode's link count.
842 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
844 kprintf("hammer_mkdir (add) error %d\n", error);
850 hammer_rel_inode(nip, 0);
851 hammer_abort_transaction(&trans);
854 hammer_commit_transaction(&trans);
855 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
856 hammer_rel_inode(nip, 0);
858 cache_setunresolved(ap->a_nch);
859 cache_setvp(ap->a_nch, *ap->a_vpp);
866 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
868 * The operating system has already ensured that the directory entry
869 * does not exist and done all appropriate namespace locking.
873 hammer_vop_nmknod(struct vop_nmknod_args *ap)
875 struct hammer_transaction trans;
876 struct hammer_inode *dip;
877 struct hammer_inode *nip;
878 struct nchandle *nch;
882 dip = VTOI(ap->a_dvp);
884 if (dip->flags & HAMMER_INODE_RO)
888 * Create a transaction to cover the operations we perform.
890 hammer_start_transaction(&trans, dip->hmp);
893 * Create a new filesystem object of the requested type. The
894 * returned inode will be referenced but not locked.
896 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
898 hammer_abort_transaction(&trans);
904 * Add the new filesystem object to the directory. This will also
905 * bump the inode's link count.
907 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
913 hammer_rel_inode(nip, 0);
914 hammer_abort_transaction(&trans);
917 hammer_commit_transaction(&trans);
918 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
919 hammer_rel_inode(nip, 0);
921 cache_setunresolved(ap->a_nch);
922 cache_setvp(ap->a_nch, *ap->a_vpp);
929 * hammer_vop_open { vp, mode, cred, fp }
933 hammer_vop_open(struct vop_open_args *ap)
935 if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
938 return(vop_stdopen(ap));
942 * hammer_vop_pathconf { vp, name, retval }
946 hammer_vop_pathconf(struct vop_pathconf_args *ap)
952 * hammer_vop_print { vp }
956 hammer_vop_print(struct vop_print_args *ap)
962 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
966 hammer_vop_readdir(struct vop_readdir_args *ap)
968 struct hammer_cursor cursor;
969 struct hammer_inode *ip;
971 hammer_record_ondisk_t rec;
972 hammer_base_elm_t base;
982 saveoff = uio->uio_offset;
984 if (ap->a_ncookies) {
985 ncookies = uio->uio_resid / 16 + 1;
988 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
997 * Handle artificial entries
1001 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1005 cookies[cookie_index] = saveoff;
1008 if (cookie_index == ncookies)
1012 if (ip->ino_data.parent_obj_id) {
1013 r = vop_write_dirent(&error, uio,
1014 ip->ino_data.parent_obj_id,
1017 r = vop_write_dirent(&error, uio,
1018 ip->obj_id, DT_DIR, 2, "..");
1023 cookies[cookie_index] = saveoff;
1026 if (cookie_index == ncookies)
1031 * Key range (begin and end inclusive) to scan. Directory keys
1032 * directly translate to a 64 bit 'seek' position.
1034 hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1035 cursor.key_beg.obj_id = ip->obj_id;
1036 cursor.key_beg.create_tid = 0;
1037 cursor.key_beg.delete_tid = 0;
1038 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1039 cursor.key_beg.obj_type = 0;
1040 cursor.key_beg.key = saveoff;
1042 cursor.key_end = cursor.key_beg;
1043 cursor.key_end.key = HAMMER_MAX_KEY;
1044 cursor.asof = ip->obj_asof;
1045 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1047 error = hammer_ip_first(&cursor, ip);
1049 while (error == 0) {
1050 error = hammer_ip_resolve_record_and_data(&cursor);
1053 rec = cursor.record;
1054 base = &rec->base.base;
1055 saveoff = base->key;
1057 if (base->obj_id != ip->obj_id)
1058 panic("readdir: bad record at %p", cursor.node);
1060 r = vop_write_dirent(
1061 &error, uio, rec->entry.obj_id,
1062 hammer_get_dtype(rec->entry.base.base.obj_type),
1063 rec->entry.base.data_len,
1064 (void *)cursor.data);
1069 cookies[cookie_index] = base->key;
1071 if (cookie_index == ncookies)
1073 error = hammer_ip_next(&cursor);
1075 hammer_done_cursor(&cursor);
1079 *ap->a_eofflag = (error == ENOENT);
1080 uio->uio_offset = saveoff;
1081 if (error && cookie_index == 0) {
1082 if (error == ENOENT)
1085 kfree(cookies, M_TEMP);
1086 *ap->a_ncookies = 0;
1087 *ap->a_cookies = NULL;
1090 if (error == ENOENT)
1093 *ap->a_ncookies = cookie_index;
1094 *ap->a_cookies = cookies;
1101 * hammer_vop_readlink { vp, uio, cred }
1105 hammer_vop_readlink(struct vop_readlink_args *ap)
1107 struct hammer_cursor cursor;
1108 struct hammer_inode *ip;
1111 ip = VTOI(ap->a_vp);
1112 hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1115 * Key range (begin and end inclusive) to scan. Directory keys
1116 * directly translate to a 64 bit 'seek' position.
1118 cursor.key_beg.obj_id = ip->obj_id;
1119 cursor.key_beg.create_tid = 0;
1120 cursor.key_beg.delete_tid = 0;
1121 cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1122 cursor.key_beg.obj_type = 0;
1123 cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1124 cursor.asof = ip->obj_asof;
1125 cursor.flags |= HAMMER_CURSOR_ASOF;
1127 error = hammer_ip_lookup(&cursor, ip);
1129 error = hammer_ip_resolve_data(&cursor);
1131 error = uiomove((char *)cursor.data,
1132 cursor.record->base.data_len,
1136 hammer_done_cursor(&cursor);
1141 * hammer_vop_nremove { nch, dvp, cred }
1145 hammer_vop_nremove(struct vop_nremove_args *ap)
1147 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1151 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1155 hammer_vop_nrename(struct vop_nrename_args *ap)
1157 struct hammer_transaction trans;
1158 struct namecache *fncp;
1159 struct namecache *tncp;
1160 struct hammer_inode *fdip;
1161 struct hammer_inode *tdip;
1162 struct hammer_inode *ip;
1163 struct hammer_cursor cursor;
1164 union hammer_record_ondisk *rec;
1168 fdip = VTOI(ap->a_fdvp);
1169 tdip = VTOI(ap->a_tdvp);
1170 fncp = ap->a_fnch->ncp;
1171 tncp = ap->a_tnch->ncp;
1172 ip = VTOI(fncp->nc_vp);
1173 KKASSERT(ip != NULL);
1175 if (fdip->flags & HAMMER_INODE_RO)
1177 if (tdip->flags & HAMMER_INODE_RO)
1179 if (ip->flags & HAMMER_INODE_RO)
1182 hammer_start_transaction(&trans, fdip->hmp);
1185 * Remove tncp from the target directory and then link ip as
1186 * tncp. XXX pass trans to dounlink
1188 * Force the inode sync-time to match the transaction so it is
1189 * in-sync with the creation of the target directory entry.
1191 error = hammer_dounlink(ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1192 if (error == 0 || error == ENOENT) {
1193 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1195 ip->ino_data.parent_obj_id = tdip->obj_id;
1196 hammer_modify_inode(&trans, ip,
1197 HAMMER_INODE_DDIRTY | HAMMER_INODE_TIDLOCKED);
1201 goto failed; /* XXX */
1204 * Locate the record in the originating directory and remove it.
1206 * Calculate the namekey and setup the key range for the scan. This
1207 * works kinda like a chained hash table where the lower 32 bits
1208 * of the namekey synthesize the chain.
1210 * The key range is inclusive of both key_beg and key_end.
1212 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1214 hammer_init_cursor_hmp(&cursor, &fdip->cache[0], fdip->hmp);
1215 cursor.key_beg.obj_id = fdip->obj_id;
1216 cursor.key_beg.key = namekey;
1217 cursor.key_beg.create_tid = 0;
1218 cursor.key_beg.delete_tid = 0;
1219 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1220 cursor.key_beg.obj_type = 0;
1222 cursor.key_end = cursor.key_beg;
1223 cursor.key_end.key |= 0xFFFFFFFFULL;
1224 cursor.asof = fdip->obj_asof;
1225 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1228 * Scan all matching records (the chain), locate the one matching
1229 * the requested path component.
1231 * The hammer_ip_*() functions merge in-memory records with on-disk
1232 * records for the purposes of the search.
1234 error = hammer_ip_first(&cursor, fdip);
1235 while (error == 0) {
1236 if (hammer_ip_resolve_data(&cursor) != 0)
1238 rec = cursor.record;
1239 if (fncp->nc_nlen == rec->entry.base.data_len &&
1240 bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1243 error = hammer_ip_next(&cursor);
1247 * If all is ok we have to get the inode so we can adjust nlinks.
1249 * WARNING: hammer_ip_del_directory() may have to terminate the
1250 * cursor to avoid a recursion. It's ok to call hammer_done_cursor()
1254 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1255 hammer_done_cursor(&cursor);
1257 cache_rename(ap->a_fnch, ap->a_tnch);
1258 if (error == EDEADLK)
1262 hammer_commit_transaction(&trans);
1264 hammer_abort_transaction(&trans);
1270 * hammer_vop_nrmdir { nch, dvp, cred }
1274 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1276 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1280 * hammer_vop_setattr { vp, vap, cred }
1284 hammer_vop_setattr(struct vop_setattr_args *ap)
1286 struct hammer_transaction trans;
1288 struct hammer_inode *ip;
1292 int64_t aligned_size;
1297 ip = ap->a_vp->v_data;
1300 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1302 if (ip->flags & HAMMER_INODE_RO)
1305 hammer_start_transaction(&trans, ip->hmp);
1308 if (vap->va_flags != VNOVAL) {
1309 flags = ip->ino_data.uflags;
1310 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1311 hammer_to_unix_xid(&ip->ino_data.uid),
1314 if (ip->ino_data.uflags != flags) {
1315 ip->ino_data.uflags = flags;
1316 modflags |= HAMMER_INODE_DDIRTY;
1318 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1325 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1329 if (vap->va_uid != (uid_t)VNOVAL) {
1330 hammer_guid_to_uuid(&uuid, vap->va_uid);
1331 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1332 ip->ino_data.uid = uuid;
1333 modflags |= HAMMER_INODE_DDIRTY;
1336 if (vap->va_gid != (uid_t)VNOVAL) {
1337 hammer_guid_to_uuid(&uuid, vap->va_gid);
1338 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1339 ip->ino_data.gid = uuid;
1340 modflags |= HAMMER_INODE_DDIRTY;
1343 while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1344 switch(ap->a_vp->v_type) {
1346 if (vap->va_size == ip->ino_rec.ino_size)
1348 if (vap->va_size < ip->ino_rec.ino_size) {
1349 vtruncbuf(ap->a_vp, vap->va_size,
1353 vnode_pager_setsize(ap->a_vp, vap->va_size);
1356 ip->ino_rec.ino_size = vap->va_size;
1357 modflags |= HAMMER_INODE_RDIRTY;
1358 aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1359 ~(int64_t)HAMMER_BUFMASK;
1362 error = hammer_ip_delete_range(&trans, ip,
1364 0x7FFFFFFFFFFFFFFFLL);
1367 * If truncating we have to clean out a portion of
1368 * the last block on-disk.
1370 if (truncating && error == 0 &&
1371 vap->va_size < aligned_size) {
1375 offset = vap->va_size & HAMMER_BUFMASK;
1376 error = bread(ap->a_vp,
1377 aligned_size - HAMMER_BUFSIZE,
1378 HAMMER_BUFSIZE, &bp);
1380 bzero(bp->b_data + offset,
1381 HAMMER_BUFSIZE - offset);
1389 error = hammer_ip_delete_range(&trans, ip,
1391 0x7FFFFFFFFFFFFFFFLL);
1392 ip->ino_rec.ino_size = vap->va_size;
1393 modflags |= HAMMER_INODE_RDIRTY;
1401 if (vap->va_atime.tv_sec != VNOVAL) {
1402 ip->ino_rec.ino_atime =
1403 hammer_timespec_to_transid(&vap->va_atime);
1404 modflags |= HAMMER_INODE_ITIMES;
1406 if (vap->va_mtime.tv_sec != VNOVAL) {
1407 ip->ino_rec.ino_mtime =
1408 hammer_timespec_to_transid(&vap->va_mtime);
1409 modflags |= HAMMER_INODE_ITIMES;
1411 if (vap->va_mode != (mode_t)VNOVAL) {
1412 if (ip->ino_data.mode != vap->va_mode) {
1413 ip->ino_data.mode = vap->va_mode;
1414 modflags |= HAMMER_INODE_DDIRTY;
1419 hammer_abort_transaction(&trans);
1421 hammer_modify_inode(&trans, ip, modflags);
1422 hammer_commit_transaction(&trans);
1428 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1432 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1434 struct hammer_transaction trans;
1435 struct hammer_inode *dip;
1436 struct hammer_inode *nip;
1437 struct nchandle *nch;
1438 hammer_record_t record;
1442 ap->a_vap->va_type = VLNK;
1445 dip = VTOI(ap->a_dvp);
1447 if (dip->flags & HAMMER_INODE_RO)
1451 * Create a transaction to cover the operations we perform.
1453 hammer_start_transaction(&trans, dip->hmp);
1456 * Create a new filesystem object of the requested type. The
1457 * returned inode will be referenced but not locked.
1460 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1462 hammer_abort_transaction(&trans);
1468 * Add the new filesystem object to the directory. This will also
1469 * bump the inode's link count.
1471 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1474 * Add a record representing the symlink. symlink stores the link
1475 * as pure data, not a string, and is no \0 terminated.
1478 record = hammer_alloc_mem_record(nip);
1479 bytes = strlen(ap->a_target);
1481 record->rec.base.base.key = HAMMER_FIXKEY_SYMLINK;
1482 record->rec.base.base.rec_type = HAMMER_RECTYPE_FIX;
1483 record->rec.base.data_len = bytes;
1484 record->data = (void *)ap->a_target;
1485 /* will be reallocated by routine below */
1486 error = hammer_ip_add_record(&trans, record);
1489 * Set the file size to the length of the link.
1492 nip->ino_rec.ino_size = bytes;
1493 hammer_modify_inode(&trans, nip, HAMMER_INODE_RDIRTY);
1501 hammer_rel_inode(nip, 0);
1502 hammer_abort_transaction(&trans);
1505 hammer_commit_transaction(&trans);
1506 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1507 hammer_rel_inode(nip, 0);
1509 cache_setunresolved(ap->a_nch);
1510 cache_setvp(ap->a_nch, *ap->a_vpp);
1517 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1521 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1523 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1527 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1531 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1533 struct hammer_inode *ip = ap->a_vp->v_data;
1535 return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1536 ap->a_fflag, ap->a_cred));
1541 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1546 mp = ap->a_head.a_ops->head.vv_mount;
1549 case MOUNTCTL_SET_EXPORT:
1550 if (ap->a_ctllen != sizeof(struct export_args))
1552 error = hammer_vfs_export(mp, ap->a_op,
1553 (const struct export_args *)ap->a_ctl);
1556 error = journal_mountctl(ap);
1563 * hammer_vop_strategy { vp, bio }
1565 * Strategy call, used for regular file read & write only. Note that the
1566 * bp may represent a cluster.
1568 * To simplify operation and allow better optimizations in the future,
1569 * this code does not make any assumptions with regards to buffer alignment
1574 hammer_vop_strategy(struct vop_strategy_args *ap)
1579 bp = ap->a_bio->bio_buf;
1583 error = hammer_vop_strategy_read(ap);
1586 error = hammer_vop_strategy_write(ap);
1592 bp->b_error = error;
1594 bp->b_flags |= B_ERROR;
1600 * Read from a regular file. Iterate the related records and fill in the
1601 * BIO/BUF. Gaps are zero-filled.
1603 * The support code in hammer_object.c should be used to deal with mixed
1604 * in-memory and on-disk records.
1610 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1612 struct hammer_inode *ip = ap->a_vp->v_data;
1613 struct hammer_cursor cursor;
1614 hammer_record_ondisk_t rec;
1615 hammer_base_elm_t base;
1629 hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1632 * Key range (begin and end inclusive) to scan. Note that the key's
1633 * stored in the actual records represent BASE+LEN, not BASE. The
1634 * first record containing bio_offset will have a key > bio_offset.
1636 cursor.key_beg.obj_id = ip->obj_id;
1637 cursor.key_beg.create_tid = 0;
1638 cursor.key_beg.delete_tid = 0;
1639 cursor.key_beg.obj_type = 0;
1640 cursor.key_beg.key = bio->bio_offset + 1;
1641 cursor.asof = ip->obj_asof;
1642 cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1644 cursor.key_end = cursor.key_beg;
1645 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1646 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1647 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1648 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1650 ran_end = bio->bio_offset + bp->b_bufsize;
1651 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1652 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1653 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
1654 if (tmp64 < ran_end)
1655 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1657 cursor.key_end.key = ran_end + MAXPHYS + 1;
1659 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1661 error = hammer_ip_first(&cursor, ip);
1664 while (error == 0) {
1665 error = hammer_ip_resolve_data(&cursor);
1668 rec = cursor.record;
1669 base = &rec->base.base;
1671 rec_offset = base->key - rec->data.base.data_len;
1674 * Calculate the gap, if any, and zero-fill it.
1676 n = (int)(rec_offset - (bio->bio_offset + boff));
1678 if (n > bp->b_bufsize - boff)
1679 n = bp->b_bufsize - boff;
1680 bzero((char *)bp->b_data + boff, n);
1686 * Calculate the data offset in the record and the number
1687 * of bytes we can copy.
1689 * Note there is a degenerate case here where boff may
1690 * already be at bp->b_bufsize.
1693 n = rec->data.base.data_len - roff;
1695 if (n > bp->b_bufsize - boff)
1696 n = bp->b_bufsize - boff;
1697 bcopy((char *)cursor.data + roff,
1698 (char *)bp->b_data + boff, n);
1700 if (boff == bp->b_bufsize)
1702 error = hammer_ip_next(&cursor);
1704 hammer_done_cursor(&cursor);
1707 * There may have been a gap after the last record
1709 if (error == ENOENT)
1711 if (error == 0 && boff != bp->b_bufsize) {
1712 KKASSERT(boff < bp->b_bufsize);
1713 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1714 /* boff = bp->b_bufsize; */
1721 * Write to a regular file. Iterate the related records and mark for
1722 * deletion. If existing edge records (left and right side) overlap our
1723 * write they have to be marked deleted and new records created, usually
1724 * referencing a portion of the original data. Then add a record to
1725 * represent the buffer.
1727 * The support code in hammer_object.c should be used to deal with mixed
1728 * in-memory and on-disk records.
1732 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1734 struct hammer_transaction trans;
1742 ip = ap->a_vp->v_data;
1744 if (ip->flags & HAMMER_INODE_RO)
1748 * Start a transaction using the TID stored with the bp.
1750 KKASSERT(bp->b_tid != 0);
1751 hammer_start_transaction_tid(&trans, ip->hmp, bp->b_tid);
1754 * Delete any records overlapping our range. This function will
1755 * (eventually) properly truncate partial overlaps.
1757 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1758 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1761 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1767 * Add a single record to cover the write
1770 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1771 bp->b_data, bp->b_bufsize);
1775 * If an error occured abort the transaction
1778 /* XXX undo deletion */
1779 hammer_abort_transaction(&trans);
1780 bp->b_resid = bp->b_bufsize;
1782 hammer_commit_transaction(&trans);
1790 * dounlink - disconnect a directory entry
1792 * XXX whiteout support not really in yet
1795 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1798 struct hammer_transaction trans;
1799 struct namecache *ncp;
1802 hammer_record_ondisk_t rec;
1803 struct hammer_cursor cursor;
1808 * Calculate the namekey and setup the key range for the scan. This
1809 * works kinda like a chained hash table where the lower 32 bits
1810 * of the namekey synthesize the chain.
1812 * The key range is inclusive of both key_beg and key_end.
1817 if (dip->flags & HAMMER_INODE_RO)
1820 hammer_start_transaction(&trans, dip->hmp);
1822 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1824 hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
1825 cursor.key_beg.obj_id = dip->obj_id;
1826 cursor.key_beg.key = namekey;
1827 cursor.key_beg.create_tid = 0;
1828 cursor.key_beg.delete_tid = 0;
1829 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1830 cursor.key_beg.obj_type = 0;
1832 cursor.key_end = cursor.key_beg;
1833 cursor.key_end.key |= 0xFFFFFFFFULL;
1834 cursor.asof = dip->obj_asof;
1835 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1838 * Scan all matching records (the chain), locate the one matching
1839 * the requested path component. info->last_error contains the
1840 * error code on search termination and could be 0, ENOENT, or
1843 * The hammer_ip_*() functions merge in-memory records with on-disk
1844 * records for the purposes of the search.
1846 error = hammer_ip_first(&cursor, dip);
1847 while (error == 0) {
1848 error = hammer_ip_resolve_data(&cursor);
1851 rec = cursor.record;
1852 if (ncp->nc_nlen == rec->entry.base.data_len &&
1853 bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1856 error = hammer_ip_next(&cursor);
1860 * If all is ok we have to get the inode so we can adjust nlinks.
1862 * If the target is a directory, it must be empty.
1865 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
1867 dip->hmp->asof, 0, &error);
1868 if (error == ENOENT) {
1869 kprintf("obj_id %016llx\n", rec->entry.obj_id);
1870 Debugger("ENOENT unlinking object that should exist, cont to sync");
1871 hammer_sync_hmp(dip->hmp, MNT_NOWAIT);
1872 Debugger("ENOENT - sync done");
1874 if (error == 0 && ip->ino_rec.base.base.obj_type ==
1875 HAMMER_OBJTYPE_DIRECTORY) {
1876 error = hammer_ip_check_directory_empty(&trans, ip);
1879 * WARNING: hammer_ip_del_directory() may have to terminate
1880 * the cursor to avoid a lock recursion. It's ok to call
1881 * hammer_done_cursor() twice.
1884 error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1886 cache_setunresolved(nch);
1887 cache_setvp(nch, NULL);
1890 cache_inval_vp(ip->vp, CINV_DESTROY);
1892 hammer_rel_inode(ip, 0);
1894 hammer_done_cursor(&cursor);
1895 if (error == EDEADLK)
1899 hammer_commit_transaction(&trans);
1901 hammer_abort_transaction(&trans);
1905 /************************************************************************
1906 * FIFO AND SPECFS OPS *
1907 ************************************************************************
1912 hammer_vop_fifoclose (struct vop_close_args *ap)
1914 /* XXX update itimes */
1915 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1919 hammer_vop_fiforead (struct vop_read_args *ap)
1923 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1924 /* XXX update access time */
1929 hammer_vop_fifowrite (struct vop_write_args *ap)
1933 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1934 /* XXX update access time */
1939 hammer_vop_specclose (struct vop_close_args *ap)
1941 /* XXX update itimes */
1942 return (VOCALL(&spec_vnode_vops, &ap->a_head));
1946 hammer_vop_specread (struct vop_read_args *ap)
1948 /* XXX update access time */
1949 return (VOCALL(&spec_vnode_vops, &ap->a_head));
1953 hammer_vop_specwrite (struct vop_write_args *ap)
1955 /* XXX update last change time */
1956 return (VOCALL(&spec_vnode_vops, &ap->a_head));